Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License version 2 as published by | |
8 | * the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., 51 | |
17 | * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | * | |
19 | * Authors: Adrian Hunter | |
20 | * Artem Bityutskiy (Битюцкий Артём) | |
21 | */ | |
22 | ||
23 | /* | |
24 | * This file implements functions needed to recover from unclean un-mounts. | |
25 | * When UBIFS is mounted, it checks a flag on the master node to determine if | |
af901ca1 | 26 | * an un-mount was completed successfully. If not, the process of mounting |
6fb4374f | 27 | * incorporates additional checking and fixing of on-flash data structures. |
1e51764a AB |
28 | * UBIFS always cleans away all remnants of an unclean un-mount, so that |
29 | * errors do not accumulate. However UBIFS defers recovery if it is mounted | |
30 | * read-only, and the flash is not modified in that case. | |
31 | */ | |
32 | ||
33 | #include <linux/crc32.h> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
1e51764a AB |
35 | #include "ubifs.h" |
36 | ||
37 | /** | |
38 | * is_empty - determine whether a buffer is empty (contains all 0xff). | |
39 | * @buf: buffer to clean | |
40 | * @len: length of buffer | |
41 | * | |
42 | * This function returns %1 if the buffer is empty (contains all 0xff) otherwise | |
43 | * %0 is returned. | |
44 | */ | |
45 | static int is_empty(void *buf, int len) | |
46 | { | |
47 | uint8_t *p = buf; | |
48 | int i; | |
49 | ||
50 | for (i = 0; i < len; i++) | |
51 | if (*p++ != 0xff) | |
52 | return 0; | |
53 | return 1; | |
54 | } | |
55 | ||
06112547 AB |
56 | /** |
57 | * first_non_ff - find offset of the first non-0xff byte. | |
58 | * @buf: buffer to search in | |
59 | * @len: length of buffer | |
60 | * | |
61 | * This function returns offset of the first non-0xff byte in @buf or %-1 if | |
62 | * the buffer contains only 0xff bytes. | |
63 | */ | |
64 | static int first_non_ff(void *buf, int len) | |
65 | { | |
66 | uint8_t *p = buf; | |
67 | int i; | |
68 | ||
69 | for (i = 0; i < len; i++) | |
70 | if (*p++ != 0xff) | |
71 | return i; | |
72 | return -1; | |
73 | } | |
74 | ||
1e51764a AB |
75 | /** |
76 | * get_master_node - get the last valid master node allowing for corruption. | |
77 | * @c: UBIFS file-system description object | |
78 | * @lnum: LEB number | |
79 | * @pbuf: buffer containing the LEB read, is returned here | |
80 | * @mst: master node, if found, is returned here | |
81 | * @cor: corruption, if found, is returned here | |
82 | * | |
83 | * This function allocates a buffer, reads the LEB into it, and finds and | |
84 | * returns the last valid master node allowing for one area of corruption. | |
85 | * The corrupt area, if there is one, must be consistent with the assumption | |
86 | * that it is the result of an unclean unmount while the master node was being | |
87 | * written. Under those circumstances, it is valid to use the previously written | |
88 | * master node. | |
89 | * | |
90 | * This function returns %0 on success and a negative error code on failure. | |
91 | */ | |
92 | static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf, | |
93 | struct ubifs_mst_node **mst, void **cor) | |
94 | { | |
95 | const int sz = c->mst_node_alsz; | |
96 | int err, offs, len; | |
97 | void *sbuf, *buf; | |
98 | ||
99 | sbuf = vmalloc(c->leb_size); | |
100 | if (!sbuf) | |
101 | return -ENOMEM; | |
102 | ||
103 | err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size); | |
104 | if (err && err != -EBADMSG) | |
105 | goto out_free; | |
106 | ||
107 | /* Find the first position that is definitely not a node */ | |
108 | offs = 0; | |
109 | buf = sbuf; | |
110 | len = c->leb_size; | |
111 | while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) { | |
112 | struct ubifs_ch *ch = buf; | |
113 | ||
114 | if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) | |
115 | break; | |
116 | offs += sz; | |
117 | buf += sz; | |
118 | len -= sz; | |
119 | } | |
120 | /* See if there was a valid master node before that */ | |
121 | if (offs) { | |
122 | int ret; | |
123 | ||
124 | offs -= sz; | |
125 | buf -= sz; | |
126 | len += sz; | |
127 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
128 | if (ret != SCANNED_A_NODE && offs) { | |
129 | /* Could have been corruption so check one place back */ | |
130 | offs -= sz; | |
131 | buf -= sz; | |
132 | len += sz; | |
133 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
134 | if (ret != SCANNED_A_NODE) | |
135 | /* | |
136 | * We accept only one area of corruption because | |
137 | * we are assuming that it was caused while | |
138 | * trying to write a master node. | |
139 | */ | |
140 | goto out_err; | |
141 | } | |
142 | if (ret == SCANNED_A_NODE) { | |
143 | struct ubifs_ch *ch = buf; | |
144 | ||
145 | if (ch->node_type != UBIFS_MST_NODE) | |
146 | goto out_err; | |
147 | dbg_rcvry("found a master node at %d:%d", lnum, offs); | |
148 | *mst = buf; | |
149 | offs += sz; | |
150 | buf += sz; | |
151 | len -= sz; | |
152 | } | |
153 | } | |
154 | /* Check for corruption */ | |
155 | if (offs < c->leb_size) { | |
156 | if (!is_empty(buf, min_t(int, len, sz))) { | |
157 | *cor = buf; | |
158 | dbg_rcvry("found corruption at %d:%d", lnum, offs); | |
159 | } | |
160 | offs += sz; | |
161 | buf += sz; | |
162 | len -= sz; | |
163 | } | |
164 | /* Check remaining empty space */ | |
165 | if (offs < c->leb_size) | |
166 | if (!is_empty(buf, len)) | |
167 | goto out_err; | |
168 | *pbuf = sbuf; | |
169 | return 0; | |
170 | ||
171 | out_err: | |
172 | err = -EINVAL; | |
173 | out_free: | |
174 | vfree(sbuf); | |
175 | *mst = NULL; | |
176 | *cor = NULL; | |
177 | return err; | |
178 | } | |
179 | ||
180 | /** | |
181 | * write_rcvrd_mst_node - write recovered master node. | |
182 | * @c: UBIFS file-system description object | |
183 | * @mst: master node | |
184 | * | |
185 | * This function returns %0 on success and a negative error code on failure. | |
186 | */ | |
187 | static int write_rcvrd_mst_node(struct ubifs_info *c, | |
188 | struct ubifs_mst_node *mst) | |
189 | { | |
190 | int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz; | |
0ecb9529 | 191 | __le32 save_flags; |
1e51764a AB |
192 | |
193 | dbg_rcvry("recovery"); | |
194 | ||
195 | save_flags = mst->flags; | |
0ecb9529 | 196 | mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY); |
1e51764a AB |
197 | |
198 | ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1); | |
199 | err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM); | |
200 | if (err) | |
201 | goto out; | |
202 | err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM); | |
203 | if (err) | |
204 | goto out; | |
205 | out: | |
206 | mst->flags = save_flags; | |
207 | return err; | |
208 | } | |
209 | ||
210 | /** | |
211 | * ubifs_recover_master_node - recover the master node. | |
212 | * @c: UBIFS file-system description object | |
213 | * | |
214 | * This function recovers the master node from corruption that may occur due to | |
215 | * an unclean unmount. | |
216 | * | |
217 | * This function returns %0 on success and a negative error code on failure. | |
218 | */ | |
219 | int ubifs_recover_master_node(struct ubifs_info *c) | |
220 | { | |
221 | void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL; | |
222 | struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst; | |
223 | const int sz = c->mst_node_alsz; | |
224 | int err, offs1, offs2; | |
225 | ||
226 | dbg_rcvry("recovery"); | |
227 | ||
228 | err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1); | |
229 | if (err) | |
230 | goto out_free; | |
231 | ||
232 | err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2); | |
233 | if (err) | |
234 | goto out_free; | |
235 | ||
236 | if (mst1) { | |
237 | offs1 = (void *)mst1 - buf1; | |
238 | if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) && | |
239 | (offs1 == 0 && !cor1)) { | |
240 | /* | |
241 | * mst1 was written by recovery at offset 0 with no | |
242 | * corruption. | |
243 | */ | |
244 | dbg_rcvry("recovery recovery"); | |
245 | mst = mst1; | |
246 | } else if (mst2) { | |
247 | offs2 = (void *)mst2 - buf2; | |
248 | if (offs1 == offs2) { | |
249 | /* Same offset, so must be the same */ | |
250 | if (memcmp((void *)mst1 + UBIFS_CH_SZ, | |
251 | (void *)mst2 + UBIFS_CH_SZ, | |
252 | UBIFS_MST_NODE_SZ - UBIFS_CH_SZ)) | |
253 | goto out_err; | |
254 | mst = mst1; | |
255 | } else if (offs2 + sz == offs1) { | |
256 | /* 1st LEB was written, 2nd was not */ | |
257 | if (cor1) | |
258 | goto out_err; | |
259 | mst = mst1; | |
260 | } else if (offs1 == 0 && offs2 + sz >= c->leb_size) { | |
261 | /* 1st LEB was unmapped and written, 2nd not */ | |
262 | if (cor1) | |
263 | goto out_err; | |
264 | mst = mst1; | |
265 | } else | |
266 | goto out_err; | |
267 | } else { | |
268 | /* | |
269 | * 2nd LEB was unmapped and about to be written, so | |
270 | * there must be only one master node in the first LEB | |
271 | * and no corruption. | |
272 | */ | |
273 | if (offs1 != 0 || cor1) | |
274 | goto out_err; | |
275 | mst = mst1; | |
276 | } | |
277 | } else { | |
278 | if (!mst2) | |
279 | goto out_err; | |
280 | /* | |
281 | * 1st LEB was unmapped and about to be written, so there must | |
282 | * be no room left in 2nd LEB. | |
283 | */ | |
284 | offs2 = (void *)mst2 - buf2; | |
285 | if (offs2 + sz + sz <= c->leb_size) | |
286 | goto out_err; | |
287 | mst = mst2; | |
288 | } | |
289 | ||
348709ba | 290 | ubifs_msg("recovered master node from LEB %d", |
1e51764a AB |
291 | (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1)); |
292 | ||
293 | memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ); | |
294 | ||
2ef13294 | 295 | if (c->ro_mount) { |
1e51764a AB |
296 | /* Read-only mode. Keep a copy for switching to rw mode */ |
297 | c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL); | |
298 | if (!c->rcvrd_mst_node) { | |
299 | err = -ENOMEM; | |
300 | goto out_free; | |
301 | } | |
302 | memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ); | |
303 | } else { | |
304 | /* Write the recovered master node */ | |
305 | c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1; | |
306 | err = write_rcvrd_mst_node(c, c->mst_node); | |
307 | if (err) | |
308 | goto out_free; | |
309 | } | |
310 | ||
311 | vfree(buf2); | |
312 | vfree(buf1); | |
313 | ||
314 | return 0; | |
315 | ||
316 | out_err: | |
317 | err = -EINVAL; | |
318 | out_free: | |
319 | ubifs_err("failed to recover master node"); | |
320 | if (mst1) { | |
321 | dbg_err("dumping first master node"); | |
322 | dbg_dump_node(c, mst1); | |
323 | } | |
324 | if (mst2) { | |
325 | dbg_err("dumping second master node"); | |
326 | dbg_dump_node(c, mst2); | |
327 | } | |
328 | vfree(buf2); | |
329 | vfree(buf1); | |
330 | return err; | |
331 | } | |
332 | ||
333 | /** | |
334 | * ubifs_write_rcvrd_mst_node - write the recovered master node. | |
335 | * @c: UBIFS file-system description object | |
336 | * | |
337 | * This function writes the master node that was recovered during mounting in | |
338 | * read-only mode and must now be written because we are remounting rw. | |
339 | * | |
340 | * This function returns %0 on success and a negative error code on failure. | |
341 | */ | |
342 | int ubifs_write_rcvrd_mst_node(struct ubifs_info *c) | |
343 | { | |
344 | int err; | |
345 | ||
346 | if (!c->rcvrd_mst_node) | |
347 | return 0; | |
348 | c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
349 | c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
350 | err = write_rcvrd_mst_node(c, c->rcvrd_mst_node); | |
351 | if (err) | |
352 | return err; | |
353 | kfree(c->rcvrd_mst_node); | |
354 | c->rcvrd_mst_node = NULL; | |
355 | return 0; | |
356 | } | |
357 | ||
358 | /** | |
359 | * is_last_write - determine if an offset was in the last write to a LEB. | |
360 | * @c: UBIFS file-system description object | |
361 | * @buf: buffer to check | |
362 | * @offs: offset to check | |
363 | * | |
364 | * This function returns %1 if @offs was in the last write to the LEB whose data | |
365 | * is in @buf, otherwise %0 is returned. The determination is made by checking | |
428ff9d2 | 366 | * for subsequent empty space starting from the next @c->min_io_size boundary. |
1e51764a AB |
367 | */ |
368 | static int is_last_write(const struct ubifs_info *c, void *buf, int offs) | |
369 | { | |
428ff9d2 | 370 | int empty_offs, check_len; |
1e51764a AB |
371 | uint8_t *p; |
372 | ||
1e51764a | 373 | /* |
428ff9d2 | 374 | * Round up to the next @c->min_io_size boundary i.e. @offs is in the |
1e51764a AB |
375 | * last wbuf written. After that should be empty space. |
376 | */ | |
377 | empty_offs = ALIGN(offs + 1, c->min_io_size); | |
378 | check_len = c->leb_size - empty_offs; | |
379 | p = buf + empty_offs - offs; | |
431102fe | 380 | return is_empty(p, check_len); |
1e51764a AB |
381 | } |
382 | ||
383 | /** | |
384 | * clean_buf - clean the data from an LEB sitting in a buffer. | |
385 | * @c: UBIFS file-system description object | |
386 | * @buf: buffer to clean | |
387 | * @lnum: LEB number to clean | |
388 | * @offs: offset from which to clean | |
389 | * @len: length of buffer | |
390 | * | |
391 | * This function pads up to the next min_io_size boundary (if there is one) and | |
392 | * sets empty space to all 0xff. @buf, @offs and @len are updated to the next | |
428ff9d2 | 393 | * @c->min_io_size boundary. |
1e51764a AB |
394 | */ |
395 | static void clean_buf(const struct ubifs_info *c, void **buf, int lnum, | |
396 | int *offs, int *len) | |
397 | { | |
398 | int empty_offs, pad_len; | |
399 | ||
400 | lnum = lnum; | |
401 | dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs); | |
402 | ||
1e51764a AB |
403 | ubifs_assert(!(*offs & 7)); |
404 | empty_offs = ALIGN(*offs, c->min_io_size); | |
405 | pad_len = empty_offs - *offs; | |
406 | ubifs_pad(c, *buf, pad_len); | |
407 | *offs += pad_len; | |
408 | *buf += pad_len; | |
409 | *len -= pad_len; | |
410 | memset(*buf, 0xff, c->leb_size - empty_offs); | |
411 | } | |
412 | ||
413 | /** | |
414 | * no_more_nodes - determine if there are no more nodes in a buffer. | |
415 | * @c: UBIFS file-system description object | |
416 | * @buf: buffer to check | |
417 | * @len: length of buffer | |
418 | * @lnum: LEB number of the LEB from which @buf was read | |
419 | * @offs: offset from which @buf was read | |
420 | * | |
de097578 AH |
421 | * This function ensures that the corrupted node at @offs is the last thing |
422 | * written to a LEB. This function returns %1 if more data is not found and | |
423 | * %0 if more data is found. | |
1e51764a AB |
424 | */ |
425 | static int no_more_nodes(const struct ubifs_info *c, void *buf, int len, | |
426 | int lnum, int offs) | |
427 | { | |
de097578 AH |
428 | struct ubifs_ch *ch = buf; |
429 | int skip, dlen = le32_to_cpu(ch->len); | |
1e51764a | 430 | |
de097578 AH |
431 | /* Check for empty space after the corrupt node's common header */ |
432 | skip = ALIGN(offs + UBIFS_CH_SZ, c->min_io_size) - offs; | |
433 | if (is_empty(buf + skip, len - skip)) | |
434 | return 1; | |
435 | /* | |
436 | * The area after the common header size is not empty, so the common | |
437 | * header must be intact. Check it. | |
438 | */ | |
439 | if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) { | |
440 | dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs); | |
441 | return 0; | |
1e51764a | 442 | } |
de097578 AH |
443 | /* Now we know the corrupt node's length we can skip over it */ |
444 | skip = ALIGN(offs + dlen, c->min_io_size) - offs; | |
445 | /* After which there should be empty space */ | |
446 | if (is_empty(buf + skip, len - skip)) | |
447 | return 1; | |
448 | dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip); | |
449 | return 0; | |
1e51764a AB |
450 | } |
451 | ||
452 | /** | |
453 | * fix_unclean_leb - fix an unclean LEB. | |
454 | * @c: UBIFS file-system description object | |
455 | * @sleb: scanned LEB information | |
456 | * @start: offset where scan started | |
457 | */ | |
458 | static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb, | |
459 | int start) | |
460 | { | |
461 | int lnum = sleb->lnum, endpt = start; | |
462 | ||
463 | /* Get the end offset of the last node we are keeping */ | |
464 | if (!list_empty(&sleb->nodes)) { | |
465 | struct ubifs_scan_node *snod; | |
466 | ||
467 | snod = list_entry(sleb->nodes.prev, | |
468 | struct ubifs_scan_node, list); | |
469 | endpt = snod->offs + snod->len; | |
470 | } | |
471 | ||
2ef13294 | 472 | if (c->ro_mount && !c->remounting_rw) { |
1e51764a AB |
473 | /* Add to recovery list */ |
474 | struct ubifs_unclean_leb *ucleb; | |
475 | ||
476 | dbg_rcvry("need to fix LEB %d start %d endpt %d", | |
477 | lnum, start, sleb->endpt); | |
478 | ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS); | |
479 | if (!ucleb) | |
480 | return -ENOMEM; | |
481 | ucleb->lnum = lnum; | |
482 | ucleb->endpt = endpt; | |
483 | list_add_tail(&ucleb->list, &c->unclean_leb_list); | |
484 | } else { | |
485 | /* Write the fixed LEB back to flash */ | |
486 | int err; | |
487 | ||
488 | dbg_rcvry("fixing LEB %d start %d endpt %d", | |
489 | lnum, start, sleb->endpt); | |
490 | if (endpt == 0) { | |
491 | err = ubifs_leb_unmap(c, lnum); | |
492 | if (err) | |
493 | return err; | |
494 | } else { | |
495 | int len = ALIGN(endpt, c->min_io_size); | |
496 | ||
497 | if (start) { | |
498 | err = ubi_read(c->ubi, lnum, sleb->buf, 0, | |
499 | start); | |
500 | if (err) | |
501 | return err; | |
502 | } | |
503 | /* Pad to min_io_size */ | |
504 | if (len > endpt) { | |
505 | int pad_len = len - ALIGN(endpt, 8); | |
506 | ||
507 | if (pad_len > 0) { | |
508 | void *buf = sleb->buf + len - pad_len; | |
509 | ||
510 | ubifs_pad(c, buf, pad_len); | |
511 | } | |
512 | } | |
513 | err = ubi_leb_change(c->ubi, lnum, sleb->buf, len, | |
514 | UBI_UNKNOWN); | |
515 | if (err) | |
516 | return err; | |
517 | } | |
518 | } | |
519 | return 0; | |
520 | } | |
521 | ||
522 | /** | |
523 | * drop_incomplete_group - drop nodes from an incomplete group. | |
524 | * @sleb: scanned LEB information | |
525 | * @offs: offset of dropped nodes is returned here | |
526 | * | |
527 | * This function returns %1 if nodes are dropped and %0 otherwise. | |
528 | */ | |
529 | static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs) | |
530 | { | |
531 | int dropped = 0; | |
532 | ||
533 | while (!list_empty(&sleb->nodes)) { | |
534 | struct ubifs_scan_node *snod; | |
535 | struct ubifs_ch *ch; | |
536 | ||
537 | snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node, | |
538 | list); | |
539 | ch = snod->node; | |
540 | if (ch->group_type != UBIFS_IN_NODE_GROUP) | |
541 | return dropped; | |
542 | dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs); | |
543 | *offs = snod->offs; | |
544 | list_del(&snod->list); | |
545 | kfree(snod); | |
546 | sleb->nodes_cnt -= 1; | |
547 | dropped = 1; | |
548 | } | |
549 | return dropped; | |
550 | } | |
551 | ||
552 | /** | |
553 | * ubifs_recover_leb - scan and recover a LEB. | |
554 | * @c: UBIFS file-system description object | |
555 | * @lnum: LEB number | |
556 | * @offs: offset | |
557 | * @sbuf: LEB-sized buffer to use | |
558 | * @grouped: nodes may be grouped for recovery | |
559 | * | |
560 | * This function does a scan of a LEB, but caters for errors that might have | |
561 | * been caused by the unclean unmount from which we are attempting to recover. | |
ed43f2f0 AB |
562 | * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is |
563 | * found, and a negative error code in case of failure. | |
1e51764a AB |
564 | */ |
565 | struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum, | |
566 | int offs, void *sbuf, int grouped) | |
567 | { | |
568 | int err, len = c->leb_size - offs, need_clean = 0, quiet = 1; | |
569 | int empty_chkd = 0, start = offs; | |
570 | struct ubifs_scan_leb *sleb; | |
571 | void *buf = sbuf + offs; | |
572 | ||
573 | dbg_rcvry("%d:%d", lnum, offs); | |
574 | ||
575 | sleb = ubifs_start_scan(c, lnum, offs, sbuf); | |
576 | if (IS_ERR(sleb)) | |
577 | return sleb; | |
578 | ||
579 | if (sleb->ecc) | |
580 | need_clean = 1; | |
581 | ||
582 | while (len >= 8) { | |
583 | int ret; | |
584 | ||
585 | dbg_scan("look at LEB %d:%d (%d bytes left)", | |
586 | lnum, offs, len); | |
587 | ||
588 | cond_resched(); | |
589 | ||
590 | /* | |
591 | * Scan quietly until there is an error from which we cannot | |
592 | * recover | |
593 | */ | |
594 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet); | |
595 | ||
596 | if (ret == SCANNED_A_NODE) { | |
597 | /* A valid node, and not a padding node */ | |
598 | struct ubifs_ch *ch = buf; | |
599 | int node_len; | |
600 | ||
601 | err = ubifs_add_snod(c, sleb, buf, offs); | |
602 | if (err) | |
603 | goto error; | |
604 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
605 | offs += node_len; | |
606 | buf += node_len; | |
607 | len -= node_len; | |
608 | continue; | |
609 | } | |
610 | ||
611 | if (ret > 0) { | |
612 | /* Padding bytes or a valid padding node */ | |
613 | offs += ret; | |
614 | buf += ret; | |
615 | len -= ret; | |
616 | continue; | |
617 | } | |
618 | ||
619 | if (ret == SCANNED_EMPTY_SPACE) { | |
620 | if (!is_empty(buf, len)) { | |
621 | if (!is_last_write(c, buf, offs)) | |
622 | break; | |
623 | clean_buf(c, &buf, lnum, &offs, &len); | |
624 | need_clean = 1; | |
625 | } | |
626 | empty_chkd = 1; | |
627 | break; | |
628 | } | |
629 | ||
630 | if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) | |
631 | if (is_last_write(c, buf, offs)) { | |
632 | clean_buf(c, &buf, lnum, &offs, &len); | |
633 | need_clean = 1; | |
634 | empty_chkd = 1; | |
635 | break; | |
636 | } | |
637 | ||
638 | if (ret == SCANNED_A_CORRUPT_NODE) | |
639 | if (no_more_nodes(c, buf, len, lnum, offs)) { | |
640 | clean_buf(c, &buf, lnum, &offs, &len); | |
641 | need_clean = 1; | |
642 | empty_chkd = 1; | |
643 | break; | |
644 | } | |
645 | ||
646 | if (quiet) { | |
647 | /* Redo the last scan but noisily */ | |
648 | quiet = 0; | |
649 | continue; | |
650 | } | |
651 | ||
652 | switch (ret) { | |
653 | case SCANNED_GARBAGE: | |
654 | dbg_err("garbage"); | |
655 | goto corrupted; | |
656 | case SCANNED_A_CORRUPT_NODE: | |
657 | case SCANNED_A_BAD_PAD_NODE: | |
658 | dbg_err("bad node"); | |
659 | goto corrupted; | |
660 | default: | |
661 | dbg_err("unknown"); | |
ed43f2f0 AB |
662 | err = -EINVAL; |
663 | goto error; | |
1e51764a AB |
664 | } |
665 | } | |
666 | ||
667 | if (!empty_chkd && !is_empty(buf, len)) { | |
668 | if (is_last_write(c, buf, offs)) { | |
669 | clean_buf(c, &buf, lnum, &offs, &len); | |
670 | need_clean = 1; | |
671 | } else { | |
06112547 AB |
672 | int corruption = first_non_ff(buf, len); |
673 | ||
674 | ubifs_err("corrupt empty space LEB %d:%d, corruption " | |
675 | "starts at %d", lnum, offs, corruption); | |
676 | /* Make sure we dump interesting non-0xFF data */ | |
677 | offs = corruption; | |
678 | buf += corruption; | |
1e51764a AB |
679 | goto corrupted; |
680 | } | |
681 | } | |
682 | ||
683 | /* Drop nodes from incomplete group */ | |
684 | if (grouped && drop_incomplete_group(sleb, &offs)) { | |
685 | buf = sbuf + offs; | |
686 | len = c->leb_size - offs; | |
687 | clean_buf(c, &buf, lnum, &offs, &len); | |
688 | need_clean = 1; | |
689 | } | |
690 | ||
691 | if (offs % c->min_io_size) { | |
692 | clean_buf(c, &buf, lnum, &offs, &len); | |
693 | need_clean = 1; | |
694 | } | |
695 | ||
696 | ubifs_end_scan(c, sleb, lnum, offs); | |
697 | ||
698 | if (need_clean) { | |
699 | err = fix_unclean_leb(c, sleb, start); | |
700 | if (err) | |
701 | goto error; | |
702 | } | |
703 | ||
704 | return sleb; | |
705 | ||
706 | corrupted: | |
707 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
708 | err = -EUCLEAN; | |
709 | error: | |
710 | ubifs_err("LEB %d scanning failed", lnum); | |
711 | ubifs_scan_destroy(sleb); | |
712 | return ERR_PTR(err); | |
713 | } | |
714 | ||
715 | /** | |
716 | * get_cs_sqnum - get commit start sequence number. | |
717 | * @c: UBIFS file-system description object | |
718 | * @lnum: LEB number of commit start node | |
719 | * @offs: offset of commit start node | |
720 | * @cs_sqnum: commit start sequence number is returned here | |
721 | * | |
722 | * This function returns %0 on success and a negative error code on failure. | |
723 | */ | |
724 | static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs, | |
725 | unsigned long long *cs_sqnum) | |
726 | { | |
727 | struct ubifs_cs_node *cs_node = NULL; | |
728 | int err, ret; | |
729 | ||
730 | dbg_rcvry("at %d:%d", lnum, offs); | |
731 | cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL); | |
732 | if (!cs_node) | |
733 | return -ENOMEM; | |
734 | if (c->leb_size - offs < UBIFS_CS_NODE_SZ) | |
735 | goto out_err; | |
736 | err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ); | |
737 | if (err && err != -EBADMSG) | |
738 | goto out_free; | |
739 | ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0); | |
740 | if (ret != SCANNED_A_NODE) { | |
741 | dbg_err("Not a valid node"); | |
742 | goto out_err; | |
743 | } | |
744 | if (cs_node->ch.node_type != UBIFS_CS_NODE) { | |
745 | dbg_err("Node a CS node, type is %d", cs_node->ch.node_type); | |
746 | goto out_err; | |
747 | } | |
748 | if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) { | |
749 | dbg_err("CS node cmt_no %llu != current cmt_no %llu", | |
750 | (unsigned long long)le64_to_cpu(cs_node->cmt_no), | |
751 | c->cmt_no); | |
752 | goto out_err; | |
753 | } | |
754 | *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum); | |
755 | dbg_rcvry("commit start sqnum %llu", *cs_sqnum); | |
756 | kfree(cs_node); | |
757 | return 0; | |
758 | ||
759 | out_err: | |
760 | err = -EINVAL; | |
761 | out_free: | |
762 | ubifs_err("failed to get CS sqnum"); | |
763 | kfree(cs_node); | |
764 | return err; | |
765 | } | |
766 | ||
767 | /** | |
768 | * ubifs_recover_log_leb - scan and recover a log LEB. | |
769 | * @c: UBIFS file-system description object | |
770 | * @lnum: LEB number | |
771 | * @offs: offset | |
772 | * @sbuf: LEB-sized buffer to use | |
773 | * | |
774 | * This function does a scan of a LEB, but caters for errors that might have | |
7d08ae3c AB |
775 | * been caused by unclean reboots from which we are attempting to recover |
776 | * (assume that only the last log LEB can be corrupted by an unclean reboot). | |
1e51764a AB |
777 | * |
778 | * This function returns %0 on success and a negative error code on failure. | |
779 | */ | |
780 | struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum, | |
781 | int offs, void *sbuf) | |
782 | { | |
783 | struct ubifs_scan_leb *sleb; | |
784 | int next_lnum; | |
785 | ||
786 | dbg_rcvry("LEB %d", lnum); | |
787 | next_lnum = lnum + 1; | |
788 | if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs) | |
789 | next_lnum = UBIFS_LOG_LNUM; | |
790 | if (next_lnum != c->ltail_lnum) { | |
791 | /* | |
792 | * We can only recover at the end of the log, so check that the | |
793 | * next log LEB is empty or out of date. | |
794 | */ | |
348709ba | 795 | sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0); |
1e51764a AB |
796 | if (IS_ERR(sleb)) |
797 | return sleb; | |
798 | if (sleb->nodes_cnt) { | |
799 | struct ubifs_scan_node *snod; | |
800 | unsigned long long cs_sqnum = c->cs_sqnum; | |
801 | ||
802 | snod = list_entry(sleb->nodes.next, | |
803 | struct ubifs_scan_node, list); | |
804 | if (cs_sqnum == 0) { | |
805 | int err; | |
806 | ||
807 | err = get_cs_sqnum(c, lnum, offs, &cs_sqnum); | |
808 | if (err) { | |
809 | ubifs_scan_destroy(sleb); | |
810 | return ERR_PTR(err); | |
811 | } | |
812 | } | |
813 | if (snod->sqnum > cs_sqnum) { | |
814 | ubifs_err("unrecoverable log corruption " | |
815 | "in LEB %d", lnum); | |
816 | ubifs_scan_destroy(sleb); | |
817 | return ERR_PTR(-EUCLEAN); | |
818 | } | |
819 | } | |
820 | ubifs_scan_destroy(sleb); | |
821 | } | |
822 | return ubifs_recover_leb(c, lnum, offs, sbuf, 0); | |
823 | } | |
824 | ||
825 | /** | |
826 | * recover_head - recover a head. | |
827 | * @c: UBIFS file-system description object | |
828 | * @lnum: LEB number of head to recover | |
829 | * @offs: offset of head to recover | |
830 | * @sbuf: LEB-sized buffer to use | |
831 | * | |
832 | * This function ensures that there is no data on the flash at a head location. | |
833 | * | |
834 | * This function returns %0 on success and a negative error code on failure. | |
835 | */ | |
836 | static int recover_head(const struct ubifs_info *c, int lnum, int offs, | |
837 | void *sbuf) | |
838 | { | |
431102fe | 839 | int len, err; |
1e51764a AB |
840 | |
841 | if (c->min_io_size > 1) | |
842 | len = c->min_io_size; | |
843 | else | |
844 | len = 512; | |
845 | if (offs + len > c->leb_size) | |
846 | len = c->leb_size - offs; | |
847 | ||
848 | if (!len) | |
849 | return 0; | |
850 | ||
851 | /* Read at the head location and check it is empty flash */ | |
852 | err = ubi_read(c->ubi, lnum, sbuf, offs, len); | |
431102fe | 853 | if (err || !is_empty(sbuf, len)) { |
1e51764a AB |
854 | dbg_rcvry("cleaning head at %d:%d", lnum, offs); |
855 | if (offs == 0) | |
856 | return ubifs_leb_unmap(c, lnum); | |
857 | err = ubi_read(c->ubi, lnum, sbuf, 0, offs); | |
858 | if (err) | |
859 | return err; | |
860 | return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN); | |
861 | } | |
862 | ||
863 | return 0; | |
864 | } | |
865 | ||
866 | /** | |
867 | * ubifs_recover_inl_heads - recover index and LPT heads. | |
868 | * @c: UBIFS file-system description object | |
869 | * @sbuf: LEB-sized buffer to use | |
870 | * | |
871 | * This function ensures that there is no data on the flash at the index and | |
872 | * LPT head locations. | |
873 | * | |
874 | * This deals with the recovery of a half-completed journal commit. UBIFS is | |
875 | * careful never to overwrite the last version of the index or the LPT. Because | |
876 | * the index and LPT are wandering trees, data from a half-completed commit will | |
877 | * not be referenced anywhere in UBIFS. The data will be either in LEBs that are | |
878 | * assumed to be empty and will be unmapped anyway before use, or in the index | |
879 | * and LPT heads. | |
880 | * | |
881 | * This function returns %0 on success and a negative error code on failure. | |
882 | */ | |
883 | int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf) | |
884 | { | |
885 | int err; | |
886 | ||
2ef13294 | 887 | ubifs_assert(!c->ro_mount || c->remounting_rw); |
1e51764a AB |
888 | |
889 | dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs); | |
890 | err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf); | |
891 | if (err) | |
892 | return err; | |
893 | ||
894 | dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs); | |
895 | err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf); | |
896 | if (err) | |
897 | return err; | |
898 | ||
899 | return 0; | |
900 | } | |
901 | ||
902 | /** | |
903 | * clean_an_unclean_leb - read and write a LEB to remove corruption. | |
904 | * @c: UBIFS file-system description object | |
905 | * @ucleb: unclean LEB information | |
906 | * @sbuf: LEB-sized buffer to use | |
907 | * | |
908 | * This function reads a LEB up to a point pre-determined by the mount recovery, | |
909 | * checks the nodes, and writes the result back to the flash, thereby cleaning | |
910 | * off any following corruption, or non-fatal ECC errors. | |
911 | * | |
912 | * This function returns %0 on success and a negative error code on failure. | |
913 | */ | |
914 | static int clean_an_unclean_leb(const struct ubifs_info *c, | |
915 | struct ubifs_unclean_leb *ucleb, void *sbuf) | |
916 | { | |
917 | int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1; | |
918 | void *buf = sbuf; | |
919 | ||
920 | dbg_rcvry("LEB %d len %d", lnum, len); | |
921 | ||
922 | if (len == 0) { | |
923 | /* Nothing to read, just unmap it */ | |
924 | err = ubifs_leb_unmap(c, lnum); | |
925 | if (err) | |
926 | return err; | |
927 | return 0; | |
928 | } | |
929 | ||
930 | err = ubi_read(c->ubi, lnum, buf, offs, len); | |
931 | if (err && err != -EBADMSG) | |
932 | return err; | |
933 | ||
934 | while (len >= 8) { | |
935 | int ret; | |
936 | ||
937 | cond_resched(); | |
938 | ||
939 | /* Scan quietly until there is an error */ | |
940 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet); | |
941 | ||
942 | if (ret == SCANNED_A_NODE) { | |
943 | /* A valid node, and not a padding node */ | |
944 | struct ubifs_ch *ch = buf; | |
945 | int node_len; | |
946 | ||
947 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
948 | offs += node_len; | |
949 | buf += node_len; | |
950 | len -= node_len; | |
951 | continue; | |
952 | } | |
953 | ||
954 | if (ret > 0) { | |
955 | /* Padding bytes or a valid padding node */ | |
956 | offs += ret; | |
957 | buf += ret; | |
958 | len -= ret; | |
959 | continue; | |
960 | } | |
961 | ||
962 | if (ret == SCANNED_EMPTY_SPACE) { | |
963 | ubifs_err("unexpected empty space at %d:%d", | |
964 | lnum, offs); | |
965 | return -EUCLEAN; | |
966 | } | |
967 | ||
968 | if (quiet) { | |
969 | /* Redo the last scan but noisily */ | |
970 | quiet = 0; | |
971 | continue; | |
972 | } | |
973 | ||
974 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
975 | return -EUCLEAN; | |
976 | } | |
977 | ||
978 | /* Pad to min_io_size */ | |
979 | len = ALIGN(ucleb->endpt, c->min_io_size); | |
980 | if (len > ucleb->endpt) { | |
981 | int pad_len = len - ALIGN(ucleb->endpt, 8); | |
982 | ||
983 | if (pad_len > 0) { | |
984 | buf = c->sbuf + len - pad_len; | |
985 | ubifs_pad(c, buf, pad_len); | |
986 | } | |
987 | } | |
988 | ||
989 | /* Write back the LEB atomically */ | |
990 | err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN); | |
991 | if (err) | |
992 | return err; | |
993 | ||
994 | dbg_rcvry("cleaned LEB %d", lnum); | |
995 | ||
996 | return 0; | |
997 | } | |
998 | ||
999 | /** | |
1000 | * ubifs_clean_lebs - clean LEBs recovered during read-only mount. | |
1001 | * @c: UBIFS file-system description object | |
1002 | * @sbuf: LEB-sized buffer to use | |
1003 | * | |
1004 | * This function cleans a LEB identified during recovery that needs to be | |
1005 | * written but was not because UBIFS was mounted read-only. This happens when | |
1006 | * remounting to read-write mode. | |
1007 | * | |
1008 | * This function returns %0 on success and a negative error code on failure. | |
1009 | */ | |
1010 | int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf) | |
1011 | { | |
1012 | dbg_rcvry("recovery"); | |
1013 | while (!list_empty(&c->unclean_leb_list)) { | |
1014 | struct ubifs_unclean_leb *ucleb; | |
1015 | int err; | |
1016 | ||
1017 | ucleb = list_entry(c->unclean_leb_list.next, | |
1018 | struct ubifs_unclean_leb, list); | |
1019 | err = clean_an_unclean_leb(c, ucleb, sbuf); | |
1020 | if (err) | |
1021 | return err; | |
1022 | list_del(&ucleb->list); | |
1023 | kfree(ucleb); | |
1024 | } | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | /** | |
1029 | * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit. | |
1030 | * @c: UBIFS file-system description object | |
1031 | * | |
1032 | * Out-of-place garbage collection requires always one empty LEB with which to | |
1033 | * start garbage collection. The LEB number is recorded in c->gc_lnum and is | |
1034 | * written to the master node on unmounting. In the case of an unclean unmount | |
1035 | * the value of gc_lnum recorded in the master node is out of date and cannot | |
1036 | * be used. Instead, recovery must allocate an empty LEB for this purpose. | |
1037 | * However, there may not be enough empty space, in which case it must be | |
1038 | * possible to GC the dirtiest LEB into the GC head LEB. | |
1039 | * | |
1040 | * This function also runs the commit which causes the TNC updates from | |
1041 | * size-recovery and orphans to be written to the flash. That is important to | |
1042 | * ensure correct replay order for subsequent mounts. | |
1043 | * | |
1044 | * This function returns %0 on success and a negative error code on failure. | |
1045 | */ | |
1046 | int ubifs_rcvry_gc_commit(struct ubifs_info *c) | |
1047 | { | |
1048 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
1049 | struct ubifs_lprops lp; | |
1050 | int lnum, err; | |
1051 | ||
1052 | c->gc_lnum = -1; | |
1053 | if (wbuf->lnum == -1) { | |
1054 | dbg_rcvry("no GC head LEB"); | |
1055 | goto find_free; | |
1056 | } | |
1057 | /* | |
1058 | * See whether the used space in the dirtiest LEB fits in the GC head | |
1059 | * LEB. | |
1060 | */ | |
1061 | if (wbuf->offs == c->leb_size) { | |
1062 | dbg_rcvry("no room in GC head LEB"); | |
1063 | goto find_free; | |
1064 | } | |
1065 | err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2); | |
1066 | if (err) { | |
6fb4374f AB |
1067 | /* |
1068 | * There are no dirty or empty LEBs subject to here being | |
1069 | * enough for the index. Try to use | |
1070 | * 'ubifs_find_free_leb_for_idx()', which will return any empty | |
1071 | * LEBs (ignoring index requirements). If the index then | |
1072 | * doesn't have enough LEBs the recovery commit will fail - | |
1073 | * which is the same result anyway i.e. recovery fails. So | |
1074 | * there is no problem ignoring index requirements and just | |
1075 | * grabbing a free LEB since we have already established there | |
1076 | * is not a dirty LEB we could have used instead. | |
1077 | */ | |
1078 | if (err == -ENOSPC) { | |
1079 | dbg_rcvry("could not find a dirty LEB"); | |
1080 | goto find_free; | |
1081 | } | |
1e51764a AB |
1082 | return err; |
1083 | } | |
1084 | ubifs_assert(!(lp.flags & LPROPS_INDEX)); | |
1085 | lnum = lp.lnum; | |
1086 | if (lp.free + lp.dirty == c->leb_size) { | |
1087 | /* An empty LEB was returned */ | |
1088 | if (lp.free != c->leb_size) { | |
1089 | err = ubifs_change_one_lp(c, lnum, c->leb_size, | |
1090 | 0, 0, 0, 0); | |
1091 | if (err) | |
1092 | return err; | |
1093 | } | |
1094 | err = ubifs_leb_unmap(c, lnum); | |
1095 | if (err) | |
1096 | return err; | |
1097 | c->gc_lnum = lnum; | |
1098 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1099 | /* Run the commit */ | |
1100 | dbg_rcvry("committing"); | |
1101 | return ubifs_run_commit(c); | |
1102 | } | |
1103 | /* | |
1104 | * There was no empty LEB so the used space in the dirtiest LEB must fit | |
1105 | * in the GC head LEB. | |
1106 | */ | |
1107 | if (lp.free + lp.dirty < wbuf->offs) { | |
1108 | dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d", | |
1109 | lnum, wbuf->lnum, wbuf->offs); | |
1110 | err = ubifs_return_leb(c, lnum); | |
1111 | if (err) | |
1112 | return err; | |
1113 | goto find_free; | |
1114 | } | |
1115 | /* | |
1116 | * We run the commit before garbage collection otherwise subsequent | |
1117 | * mounts will see the GC and orphan deletion in a different order. | |
1118 | */ | |
1119 | dbg_rcvry("committing"); | |
1120 | err = ubifs_run_commit(c); | |
1121 | if (err) | |
1122 | return err; | |
1123 | /* | |
1124 | * The data in the dirtiest LEB fits in the GC head LEB, so do the GC | |
1125 | * - use locking to keep 'ubifs_assert()' happy. | |
1126 | */ | |
1127 | dbg_rcvry("GC'ing LEB %d", lnum); | |
1128 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
1129 | err = ubifs_garbage_collect_leb(c, &lp); | |
1130 | if (err >= 0) { | |
1131 | int err2 = ubifs_wbuf_sync_nolock(wbuf); | |
1132 | ||
1133 | if (err2) | |
1134 | err = err2; | |
1135 | } | |
1136 | mutex_unlock(&wbuf->io_mutex); | |
1137 | if (err < 0) { | |
1138 | dbg_err("GC failed, error %d", err); | |
1139 | if (err == -EAGAIN) | |
1140 | err = -EINVAL; | |
1141 | return err; | |
1142 | } | |
1143 | if (err != LEB_RETAINED) { | |
1144 | dbg_err("GC returned %d", err); | |
1145 | return -EINVAL; | |
1146 | } | |
1147 | err = ubifs_leb_unmap(c, c->gc_lnum); | |
1148 | if (err) | |
1149 | return err; | |
1150 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1151 | return 0; | |
1152 | ||
1153 | find_free: | |
1154 | /* | |
1155 | * There is no GC head LEB or the free space in the GC head LEB is too | |
6fb4374f AB |
1156 | * small, or there are not dirty LEBs. Allocate gc_lnum by calling |
1157 | * 'ubifs_find_free_leb_for_idx()' so GC is not run. | |
1e51764a AB |
1158 | */ |
1159 | lnum = ubifs_find_free_leb_for_idx(c); | |
1160 | if (lnum < 0) { | |
1161 | dbg_err("could not find an empty LEB"); | |
1162 | return lnum; | |
1163 | } | |
1164 | /* And reset the index flag */ | |
1165 | err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0, | |
1166 | LPROPS_INDEX, 0); | |
1167 | if (err) | |
1168 | return err; | |
1169 | c->gc_lnum = lnum; | |
1170 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1171 | /* Run the commit */ | |
1172 | dbg_rcvry("committing"); | |
1173 | return ubifs_run_commit(c); | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * struct size_entry - inode size information for recovery. | |
1178 | * @rb: link in the RB-tree of sizes | |
1179 | * @inum: inode number | |
1180 | * @i_size: size on inode | |
1181 | * @d_size: maximum size based on data nodes | |
1182 | * @exists: indicates whether the inode exists | |
1183 | * @inode: inode if pinned in memory awaiting rw mode to fix it | |
1184 | */ | |
1185 | struct size_entry { | |
1186 | struct rb_node rb; | |
1187 | ino_t inum; | |
1188 | loff_t i_size; | |
1189 | loff_t d_size; | |
1190 | int exists; | |
1191 | struct inode *inode; | |
1192 | }; | |
1193 | ||
1194 | /** | |
1195 | * add_ino - add an entry to the size tree. | |
1196 | * @c: UBIFS file-system description object | |
1197 | * @inum: inode number | |
1198 | * @i_size: size on inode | |
1199 | * @d_size: maximum size based on data nodes | |
1200 | * @exists: indicates whether the inode exists | |
1201 | */ | |
1202 | static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size, | |
1203 | loff_t d_size, int exists) | |
1204 | { | |
1205 | struct rb_node **p = &c->size_tree.rb_node, *parent = NULL; | |
1206 | struct size_entry *e; | |
1207 | ||
1208 | while (*p) { | |
1209 | parent = *p; | |
1210 | e = rb_entry(parent, struct size_entry, rb); | |
1211 | if (inum < e->inum) | |
1212 | p = &(*p)->rb_left; | |
1213 | else | |
1214 | p = &(*p)->rb_right; | |
1215 | } | |
1216 | ||
1217 | e = kzalloc(sizeof(struct size_entry), GFP_KERNEL); | |
1218 | if (!e) | |
1219 | return -ENOMEM; | |
1220 | ||
1221 | e->inum = inum; | |
1222 | e->i_size = i_size; | |
1223 | e->d_size = d_size; | |
1224 | e->exists = exists; | |
1225 | ||
1226 | rb_link_node(&e->rb, parent, p); | |
1227 | rb_insert_color(&e->rb, &c->size_tree); | |
1228 | ||
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | /** | |
1233 | * find_ino - find an entry on the size tree. | |
1234 | * @c: UBIFS file-system description object | |
1235 | * @inum: inode number | |
1236 | */ | |
1237 | static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum) | |
1238 | { | |
1239 | struct rb_node *p = c->size_tree.rb_node; | |
1240 | struct size_entry *e; | |
1241 | ||
1242 | while (p) { | |
1243 | e = rb_entry(p, struct size_entry, rb); | |
1244 | if (inum < e->inum) | |
1245 | p = p->rb_left; | |
1246 | else if (inum > e->inum) | |
1247 | p = p->rb_right; | |
1248 | else | |
1249 | return e; | |
1250 | } | |
1251 | return NULL; | |
1252 | } | |
1253 | ||
1254 | /** | |
1255 | * remove_ino - remove an entry from the size tree. | |
1256 | * @c: UBIFS file-system description object | |
1257 | * @inum: inode number | |
1258 | */ | |
1259 | static void remove_ino(struct ubifs_info *c, ino_t inum) | |
1260 | { | |
1261 | struct size_entry *e = find_ino(c, inum); | |
1262 | ||
1263 | if (!e) | |
1264 | return; | |
1265 | rb_erase(&e->rb, &c->size_tree); | |
1266 | kfree(e); | |
1267 | } | |
1268 | ||
1269 | /** | |
1270 | * ubifs_destroy_size_tree - free resources related to the size tree. | |
1271 | * @c: UBIFS file-system description object | |
1272 | */ | |
1273 | void ubifs_destroy_size_tree(struct ubifs_info *c) | |
1274 | { | |
1275 | struct rb_node *this = c->size_tree.rb_node; | |
1276 | struct size_entry *e; | |
1277 | ||
1278 | while (this) { | |
1279 | if (this->rb_left) { | |
1280 | this = this->rb_left; | |
1281 | continue; | |
1282 | } else if (this->rb_right) { | |
1283 | this = this->rb_right; | |
1284 | continue; | |
1285 | } | |
1286 | e = rb_entry(this, struct size_entry, rb); | |
1287 | if (e->inode) | |
1288 | iput(e->inode); | |
1289 | this = rb_parent(this); | |
1290 | if (this) { | |
1291 | if (this->rb_left == &e->rb) | |
1292 | this->rb_left = NULL; | |
1293 | else | |
1294 | this->rb_right = NULL; | |
1295 | } | |
1296 | kfree(e); | |
1297 | } | |
1298 | c->size_tree = RB_ROOT; | |
1299 | } | |
1300 | ||
1301 | /** | |
1302 | * ubifs_recover_size_accum - accumulate inode sizes for recovery. | |
1303 | * @c: UBIFS file-system description object | |
1304 | * @key: node key | |
1305 | * @deletion: node is for a deletion | |
1306 | * @new_size: inode size | |
1307 | * | |
1308 | * This function has two purposes: | |
1309 | * 1) to ensure there are no data nodes that fall outside the inode size | |
1310 | * 2) to ensure there are no data nodes for inodes that do not exist | |
1311 | * To accomplish those purposes, a rb-tree is constructed containing an entry | |
1312 | * for each inode number in the journal that has not been deleted, and recording | |
1313 | * the size from the inode node, the maximum size of any data node (also altered | |
1314 | * by truncations) and a flag indicating a inode number for which no inode node | |
1315 | * was present in the journal. | |
1316 | * | |
1317 | * Note that there is still the possibility that there are data nodes that have | |
1318 | * been committed that are beyond the inode size, however the only way to find | |
1319 | * them would be to scan the entire index. Alternatively, some provision could | |
1320 | * be made to record the size of inodes at the start of commit, which would seem | |
1321 | * very cumbersome for a scenario that is quite unlikely and the only negative | |
1322 | * consequence of which is wasted space. | |
1323 | * | |
1324 | * This functions returns %0 on success and a negative error code on failure. | |
1325 | */ | |
1326 | int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key, | |
1327 | int deletion, loff_t new_size) | |
1328 | { | |
1329 | ino_t inum = key_inum(c, key); | |
1330 | struct size_entry *e; | |
1331 | int err; | |
1332 | ||
1333 | switch (key_type(c, key)) { | |
1334 | case UBIFS_INO_KEY: | |
1335 | if (deletion) | |
1336 | remove_ino(c, inum); | |
1337 | else { | |
1338 | e = find_ino(c, inum); | |
1339 | if (e) { | |
1340 | e->i_size = new_size; | |
1341 | e->exists = 1; | |
1342 | } else { | |
1343 | err = add_ino(c, inum, new_size, 0, 1); | |
1344 | if (err) | |
1345 | return err; | |
1346 | } | |
1347 | } | |
1348 | break; | |
1349 | case UBIFS_DATA_KEY: | |
1350 | e = find_ino(c, inum); | |
1351 | if (e) { | |
1352 | if (new_size > e->d_size) | |
1353 | e->d_size = new_size; | |
1354 | } else { | |
1355 | err = add_ino(c, inum, 0, new_size, 0); | |
1356 | if (err) | |
1357 | return err; | |
1358 | } | |
1359 | break; | |
1360 | case UBIFS_TRUN_KEY: | |
1361 | e = find_ino(c, inum); | |
1362 | if (e) | |
1363 | e->d_size = new_size; | |
1364 | break; | |
1365 | } | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | /** | |
1370 | * fix_size_in_place - fix inode size in place on flash. | |
1371 | * @c: UBIFS file-system description object | |
1372 | * @e: inode size information for recovery | |
1373 | */ | |
1374 | static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e) | |
1375 | { | |
1376 | struct ubifs_ino_node *ino = c->sbuf; | |
1377 | unsigned char *p; | |
1378 | union ubifs_key key; | |
1379 | int err, lnum, offs, len; | |
1380 | loff_t i_size; | |
1381 | uint32_t crc; | |
1382 | ||
1383 | /* Locate the inode node LEB number and offset */ | |
1384 | ino_key_init(c, &key, e->inum); | |
1385 | err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs); | |
1386 | if (err) | |
1387 | goto out; | |
1388 | /* | |
1389 | * If the size recorded on the inode node is greater than the size that | |
1390 | * was calculated from nodes in the journal then don't change the inode. | |
1391 | */ | |
1392 | i_size = le64_to_cpu(ino->size); | |
1393 | if (i_size >= e->d_size) | |
1394 | return 0; | |
1395 | /* Read the LEB */ | |
1396 | err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size); | |
1397 | if (err) | |
1398 | goto out; | |
1399 | /* Change the size field and recalculate the CRC */ | |
1400 | ino = c->sbuf + offs; | |
1401 | ino->size = cpu_to_le64(e->d_size); | |
1402 | len = le32_to_cpu(ino->ch.len); | |
1403 | crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8); | |
1404 | ino->ch.crc = cpu_to_le32(crc); | |
1405 | /* Work out where data in the LEB ends and free space begins */ | |
1406 | p = c->sbuf; | |
1407 | len = c->leb_size - 1; | |
1408 | while (p[len] == 0xff) | |
1409 | len -= 1; | |
1410 | len = ALIGN(len + 1, c->min_io_size); | |
1411 | /* Atomically write the fixed LEB back again */ | |
1412 | err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN); | |
1413 | if (err) | |
1414 | goto out; | |
e84461ad AB |
1415 | dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ", |
1416 | (unsigned long)e->inum, lnum, offs, i_size, e->d_size); | |
1e51764a AB |
1417 | return 0; |
1418 | ||
1419 | out: | |
1420 | ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d", | |
e84461ad | 1421 | (unsigned long)e->inum, e->i_size, e->d_size, err); |
1e51764a AB |
1422 | return err; |
1423 | } | |
1424 | ||
1425 | /** | |
1426 | * ubifs_recover_size - recover inode size. | |
1427 | * @c: UBIFS file-system description object | |
1428 | * | |
1429 | * This function attempts to fix inode size discrepancies identified by the | |
1430 | * 'ubifs_recover_size_accum()' function. | |
1431 | * | |
1432 | * This functions returns %0 on success and a negative error code on failure. | |
1433 | */ | |
1434 | int ubifs_recover_size(struct ubifs_info *c) | |
1435 | { | |
1436 | struct rb_node *this = rb_first(&c->size_tree); | |
1437 | ||
1438 | while (this) { | |
1439 | struct size_entry *e; | |
1440 | int err; | |
1441 | ||
1442 | e = rb_entry(this, struct size_entry, rb); | |
1443 | if (!e->exists) { | |
1444 | union ubifs_key key; | |
1445 | ||
1446 | ino_key_init(c, &key, e->inum); | |
1447 | err = ubifs_tnc_lookup(c, &key, c->sbuf); | |
1448 | if (err && err != -ENOENT) | |
1449 | return err; | |
1450 | if (err == -ENOENT) { | |
1451 | /* Remove data nodes that have no inode */ | |
e84461ad AB |
1452 | dbg_rcvry("removing ino %lu", |
1453 | (unsigned long)e->inum); | |
1e51764a AB |
1454 | err = ubifs_tnc_remove_ino(c, e->inum); |
1455 | if (err) | |
1456 | return err; | |
1457 | } else { | |
1458 | struct ubifs_ino_node *ino = c->sbuf; | |
1459 | ||
1460 | e->exists = 1; | |
1461 | e->i_size = le64_to_cpu(ino->size); | |
1462 | } | |
1463 | } | |
1464 | if (e->exists && e->i_size < e->d_size) { | |
2ef13294 | 1465 | if (!e->inode && c->ro_mount) { |
1e51764a AB |
1466 | /* Fix the inode size and pin it in memory */ |
1467 | struct inode *inode; | |
1468 | ||
1469 | inode = ubifs_iget(c->vfs_sb, e->inum); | |
1470 | if (IS_ERR(inode)) | |
1471 | return PTR_ERR(inode); | |
1472 | if (inode->i_size < e->d_size) { | |
1473 | dbg_rcvry("ino %lu size %lld -> %lld", | |
e84461ad AB |
1474 | (unsigned long)e->inum, |
1475 | e->d_size, inode->i_size); | |
1e51764a AB |
1476 | inode->i_size = e->d_size; |
1477 | ubifs_inode(inode)->ui_size = e->d_size; | |
1478 | e->inode = inode; | |
1479 | this = rb_next(this); | |
1480 | continue; | |
1481 | } | |
1482 | iput(inode); | |
1483 | } else { | |
1484 | /* Fix the size in place */ | |
1485 | err = fix_size_in_place(c, e); | |
1486 | if (err) | |
1487 | return err; | |
1488 | if (e->inode) | |
1489 | iput(e->inode); | |
1490 | } | |
1491 | } | |
1492 | this = rb_next(this); | |
1493 | rb_erase(&e->rb, &c->size_tree); | |
1494 | kfree(e); | |
1495 | } | |
1496 | return 0; | |
1497 | } |